This Space uses MLX CPU fallback on Linux. It can verify that the model loads, but full translation on cpu-basic may exceed the request timeout. Run locally on Apple Silicon for the intended MLX GPU path.